The method freezes the model and edits eight harness components: system prompt, guardrails, memory, tool guidance, subagents, skills, scripts and plugins. For each task it runs the agent once, proposes two edits (one for general agent…
In the News: October 18, 2026 (Extra 2)
Google Cloud AI Research reports that letting an agent rewrite its own harness lifted GLM-5 from 64.2 to 75.7 on WorkBuddy-Bench.
Extra edition
Machine-readable
Download Markdown